The AI Conference returned to San Francisco on 29 September and closed on 1 October 2026. The event is now far enough behind us that the useful question has changed from “what should I watch for” to “what did the format actually deliver, and how would anyone know?” The second half of that question is harder than it sounds, because published recap material for this show is thin. This piece does not pretend otherwise. It looks at what the organiser said the event was built to do, where the public record conflicts, and how to read any AI-conference recap without getting swept up in keynote noise.

What the Organiser Built

According to the organiser’s site, the conference ran at Pier 48 in San Francisco’s Mission Rock waterfront district and was in person only. It structured the week as three days:

  • Day ZERO (29 September): workshops and a live AI Hack Day, with a cap on attendance. The organiser’s site lists the cap at 350; some third-party listings say 300.
  • Day 1 and Day 2 (30 September and 1 October): the main conference, with the organiser citing more than 5,500 attendees and more than 120 speakers.

The programme was divided into five tracks: AI Frontiers (governance, ethics and frontier research), AI Builders (shipping agentic and multi-model systems), The AI Stack (training, distributed inference and MLOps), Applied AI (real deployments, edge AI and evaluation), and AI Strategy (enterprise adoption and ROI).

That track design is the most telling part of the format. Most large AI events lead with a headline keynote and hang the rest of the schedule off it. This one split the audience by job: researchers, builders, infrastructure engineers, operators and executives each got a lane. Whether a given lane delivered is the thing to examine.

The Day ZERO Bet

A capped hands-on day sitting before the main conference is a deliberate trade. It sacrifices headcount, since a few hundred people does not move a registration total, for a smaller room where the work is practical and the speakers are reachable. For an attendee, the tell-tale sign that such a format worked is concrete: a working prototype, a contact with a practitioner you could not have reached in the main hall, or a specific tooling decision you changed.

The organiser bundled Day ZERO access into its top pass tier and lists it alongside a VIP dinner. Teams buying that tier were paying for access, not for content volume, and should judge it on access.

Where the Public Record Conflicts

We found no organiser-published recap at the time of writing. The organiser’s own site still described the 2026 show in forward-looking terms and listed testimonials only from the prior year. Several conflicts in the pre-event record are worth flagging:

  • Pricing: one listing showed a starting price of $199, another $1,295 or more. The organiser’s site shows tier names (Day ZERO three-day pass, full conference pass, team tickets) but no dollar amounts without registering. We cannot say what any individual paid.
  • Speakers: the organiser’s site names a subset of its 120-plus speakers, including Peter Norvig, Chris Lattner, Sergey Levine and Ion Stoica, but the structured data and the visible listing do not match in every case. A speaker named in one place was not always named in the other.
  • Day ZERO capacity: 300 versus 350, as above.

None of this suggests anything improper. It is the normal state of a conference’s public data a few weeks after it closes, and it is the reason a retrospective should name its sources.

How to Read an AI-Conference Recap

When recaps do appear, whether from the organiser or from attendees, a few habits cut through the noise.

Separate announcements from deployments. A keynote claim about what a model or product will do is a forecast. A track session where a team describes what they shipped, what broke and what it cost is evidence. The Applied AI and AI Builders tracks are where that second kind of material would live.

Ask what was measured. Phrases like “record attendance” and “eye-opening” are sentiment. Attendance by role, session ratings and lead volume per exhibitor are measurements. If a recap gives only the first kind, treat it as marketing.

Check the speaker list against the stage. Given the inconsistencies above, confirm that someone named in the promotion actually spoke before attributing a view to them.

Look for what was left out. If governance, evaluation or cost barely appear in a recap of a show that had a dedicated track for each, that absence says something.

For how this fits the wider pattern of AI-heavy programming, see our coverage of AI-dominated keynotes and our field guide to the modern tech conference circuit.

What to Do With It Now

If you attended, write your own recap while the sessions are fresh, and score each track against the job you went for. If you did not, the organiser has announced 12-14 October 2027 for the next edition, which gives you a full year to decide whether the track-by-job format fits your team. When you do evaluate it, ask for dollar pricing in writing and check whether Day ZERO is bundled, because the conflicting figures above suggest the headline price can differ from what teams actually pay.

Frequently Asked Questions

When and where was The AI Conference 2026?

It ran 29 September to 1 October 2026 at Pier 48 in San Francisco. The first day was a workshop and hack day, followed by two main conference days.

How big was The AI Conference?

The organiser cites more than 5,500 attendees for the main conference and more than 120 speakers across five tracks. We found no audited attendance count.

What tracks did The AI Conference offer?

The five tracks were AI Frontiers, AI Builders, The AI Stack, Applied AI and AI Strategy, each aimed at a different professional role.

When is the next edition?

The organiser’s site lists 12-14 October 2027 for the next edition.